The present invention relates generally to semiconductor diodes made from compound semiconductors or other lattice-mismatched semiconductors on silicon wafers, as well as methods of fabricating such semiconductor diodes, and more particularly to photonic applications such as light emitting diodes (LEDs), lasers, photovoltaics, and other optoelectronic uses.
This section provides background information and introduces information related to various aspects of the disclosures that are described and/or claimed below. These background statements are not admissions of prior art.
The majority of chip manufacturing takes advantage of silicon processing on high-quality, large-area, low-cost silicon wafers. Commercial manufacturers of devices made from compound semiconductors such as gallium arsenide and indium phosphide generally have been unable to take advantage of silicon wafers. They typically build light emitting diodes (LEDs), multi-junction solar cells, and other compound semiconductor devices on small, expensive wafers made of materials such as sapphire, germanium, gallium arsenide, or silicon carbide.
The challenge of making compound semiconductor devices on inexpensive substrates has widespread economic implications. Compound semiconductors are an important component of our communications infrastructure because they can emit and detect light. They are the materials in the lasers that transmit signals through optical fibers, the sensors that receive those signals, the amplifiers in cellular telephones, the amplifiers in cell phone base stations, and the circuits that transmit and receive microwave signals.
Light-emitting diodes typically consist of gallium nitride films deposited onto sapphire or silicon carbide wafers. These exotic substrates contribute to the high cost of LEDs. A sapphire wafer 4 inches in diameter typically costs around $130, and a 2-inch silicon carbide wafer can cost about $2000. By contrast, an 8-inch silicon wafer, which provides four times as much surface area as a 4-inch wafer and 16 times as much surface area as a 2-inch wafer, typically costs less than $100.
High-efficiency multi-junction solar cells typically contain layers such as germanium, gallium arsenide, and indium gallium phosphide deposited onto germanium wafers. As is the case with wafers for LEDs, germanium wafers similarly are smaller and significantly more expensive than silicon wafers.
The ability to create compound semiconductor devices on silicon wafers would facilitate market growth in several key industries.
Two key technical barriers have inhibited the practical fabrication of compound semiconductor devices on silicon wafers: the mismatch of lattice constants and the mismatch of thermal expansion coefficients.
Lattice Mismatch: In a crystal, the atoms sit in a regular periodic array known as a lattice. The distance between the atoms, known as the “lattice constant,” is typically a few angstroms (1 ångstrom=10-10 meter). Silicon has a smaller lattice constant than compound semiconductors. When compound semiconductors grow on silicon, crystalline imperfections known as misfit dislocations appear at the interface. The misfit dislocations create other crystalline defects known as threading dislocations, which propagate upward from the interface. Threading dislocations diminish the performance and the reliability of compound semiconductor devices such as lasers, solar cells, light-emitting diodes, etc.
Thermal Contraction Mismatch: Compound semiconductors typically grow at high temperatures, which can exceed 1000° C. When the wafer cools, the compound semiconductor film may contract more than the silicon wafer. As a result, the wafer may bow in a concave manner, stressing and ultimately cracking the film.
Until recently, the most promising previous efforts to grow high-quality compound semiconductors onto silicon substrates have relied on three approaches: graded buffer layers, wafer bonding, or selective growth on mesas. None of these approaches has achieved commercial success.
In graded buffer layers, the composition of the material changes gradually from substantially pure silicon to a compound semiconductor. Since the lattice constant also changes gradually, crystalline defects are less likely to form at the interface. Unfortunately, the graded buffer layers have to be relatively thick (about ten microns for a 4% lattice mismatch). The thick buffer layer increases both the costs and the likelihood of cracking.
Wafer bonding involves growing devices on expensive substrates, then lifting off the devices and bonding them to a silicon wafer. This approach rules out modern silicon processing as a route to cost reduction. Furthermore, bonding typically requires temperatures above 300° C. When the materials cool, the compound semiconductors may crack because they contract more than the silicon wafer.
Selective growth on a mesa exploits the mobility of some dislocations. The strategy is to deposit compound semiconductors in small regions (10 to 100 microns in length), thereby providing a short path where mobile dislocations can glide to the edge of the region and remove themselves from the device. However, structures created by this technique typically have a high density of threading dislocations (more than 100 million per square centimeter). This technique can't remove immobile dislocations, which predominate when the lattice mismatch exceeds 2%.
Aspect Ratio Trapping (J. S. Park et al., APL 90, 052113 (2007, hereby incorporated by reference) is a recently developed technology that makes it possible to deposit high-quality compound semiconductors, germanium or other lattice-mismatched materials on silicon wafers. FIG. 1 illustrates the principle of Aspect Ratio Trapping. We deposit a thin film of dielectric material (20) such as silicon dioxide (SiO2) or silicon nitride (SiNx) onto a silicon wafer (10). Those of ordinary skill in the art can select a variety of dielectric materials such as SiOxNy, and silicates or oxides of materials such as Hf or Zr.
We etch a trench in the dielectric material, then deposit a non-lattice-matched semiconductor (30) such as germanium or a compound semiconductor in the trench. The threading dislocations (40), shown as dotted lines, propagate upward, typically at approximately a 45 degree angle from the interface, then intersect the sidewalls of the trench, where they terminate. Threading dislocations (40) do not propagate down the length of the trench because they propagate in a direction perpendicular to the faceted growth face of the crystal—the facets guide the dislocations to the sidewalls, where the dislocations terminate. We refer to the region in the trench where the sidewalls trap threading dislocations as the “trapping region” (50). The upper region of the non-lattice-matched semiconductor (30), above the trapping region (50), is a relatively defect-free region (60).
ART addresses the issue of cracking caused from mismatch of thermal expansion coefficients for these reasons: (1) the stresses are small because the epitaxial layers are thin; (2) the material can elastically accommodate the stresses arising from thermal expansion mismatch because dimensions of the ART openings are small; and (3) the SiO2 pedestals, which are more compliant than the semiconductor materials, may deform to accommodate the stress.
Continuous high-quality films of high-quality III-V semiconductors or other non-lattice-matched materials may be provided on silicon wafers using Aspect Ratio Trapping as shown in FIG. 2. This technique is similar to the technique illustrated in FIG. 1, except that we continue to grow the non-lattice-matched semiconductor until the growth fronts from adjacent trenches coalesce to form a single continuous film (70). Additional defects, which we call “coalescence defects” (80), form at some of the “coalescence regions” where the growth fronts meet. However, the defect density is much lower than the defect density of a non-lattice-matched semiconductor grown directly on a silicon wafer.